The TBX Data Model

The TBX Data Model as an UML diagram.

The following list provides an overview of the main elements of a multilingual terminology represented in TBX format.

• TBX Resource: A TBX resource essentially represents a collection of terminological concepts (“terminological concepts”), which are represented as XML elements of type termEntry and have a unique ID. In the XML snippet in Section 4.1, there is one terminological concept with ID IATE_84. Each terminological concept is described by a set of properties, such as a subject field they belong to.
• Terminological Concept (term entry): represents a language-independent concept. Each terminological concept is associated to a Langset (see below), which can be seen as a set of language-specific Terms that express the Terminological Concept in question.
• Langset: A langset is a language-specific container for all the terms that lexicalize a “terminological concept” in a given language. The Langset contains simple terms, for which no decompositions is provided (TIG), as well as complex terms for which the decomposition information is provided (NTIG).
• Term Information Group (TIG): represents a language-specific term for which no decomposition information is provided.
• Nesting Term Information Group (NTIG): represents a language-specific term for which decomposition information is provided.
• TermGrp: contains information about a language-specific term including its morphosyntactic properties; there is one TermGrp for each TIG and NTIG
• TermCompList: represents the decomposition of a term
• TermCompGrp: represents one component of a term and its morphosyntactic properties
• DescrGrp: describes properties of a particular term, in particular different surface forms or describes contexts that document the usage of the term
• TransGrp/Transaction: contains information about a transaction that lead to the creation or modification of a term

The lemon-ontolex model

Lexical resources such as WordNet are one of the most important types of data sources for linguistic research. Such resources are complementary to another type of Web resources that contain a large amount of taxonomic data described in RDF such as DBpedia . Bridging the gap between these two types of resources means that rich linguistic information found in lexical resources (e.g., lexicons) can be used to describe information on the Web, enabling novel applications such as question answering over Linked Data . This leads to a new type of resources that is termed ontology-lexicon, which consists of an ontology describing the semantic and taxonomic nature of the domain and a lexicon describing the behaviour of the words in a language used to refer to the semantics captured by the ontology.

In that context lemon (Lexicon Model for Ontologies) was proposed to provide a de facto standard to be used by a cloud of lexical linked data resources such as WordNet, BabelNet, and UBY, among many others. The lemon model’s core consists of the following elements depicted below:

Lexical Entry.

A lexical entry, which may be a word, multiword expression or an affix, is assumed to represent a single lexical unit with common properties, especially part-of-speech, across all its forms and meanings.

Lexical Form.

A form represents a particular version of a lexical entry, for example a plural or some other inflected form. A form may have a number of representations in different orthographies (e.g., spelling variants) or media (e.g., phonetic representations).

Lexical Sense.

The sense refers to the usage of a word with a specific meaning and can also be considered as a reification of the pair of a lexical entry used with reference to a given ontology. The sense is also used as a node for the annotation of many pragmatic features such as register.

Reference.

The reference is an entity in the ontology that the entry can be interpreted as, or alternatively that can be represented by using the lexical entry.

Mapping the TBX Data Model to the ontolex-lemon model

The main data elements described above have been mapped into RDF using the above mentioned vocabularies as follows:

• TBX Resource: is not explicitly represented, the whole dataset represents the TBX resource. A TBX resource is thus represented as a void:Dataset, to which provenance and licensing information can be attached.
• Terminological Concept: is represented as a skos:Concept. The Simple Knowledge Organization System (SKOS) is a vocabulary for representing knowledge organization systems (KOS) such as thesauri, classification schemes, subject heading and taxonomies in RDF. The fundamental element of a SKOS vocabulary are concepts, defined as units of thought, ideas, meanings, or (categories of) objects and events, which underlie many knowledge organization systems. As terminologies can be seen as a special case of a knowledge organization system, using SKOS concepts to represent terminological concepts seems appropriate.
• Langset: A langset is not represented as such in the data. Instead, one ontolex:Lexicon is created for each language for which a Langset is defined. The collection of all the terms for a given language belongs to the corresponding language-specific ontolex:Lexicon
• TIG/NTIG: are represented as an ontolex:LexicalEntry; no distinction is made between terms with decomposition and terms withouth decomposition; if no decompositions information is available, this is simply omited. In that sense the representation is monotonic as the decomposition information can be added later
• TermGrp: the information about the morphosyntactic properties of a term is attached to the corresponding ontolex:LexicalEntry. The string enclosed in <term> </term> is assumed to be the ontolex:canonicalForm of the lexical entry in question.
• TermCompList: the decomposition of a term is represented using the ontolex:decomp vocabulary, creating a decomp:Component and a corresponding ontolex:LexicalEntry for each component.
• TermCompGrp: the morphosyntactic properties of a component are attached to the corresponding lexical entry that is identified (through decomp:correspondsTo) with the component in question)
• DescrGrp: descriptions of the term or context are mapped to appropriate properties of the lexical entry or the context
• TransGrp/Transaction: a transaction that creates or modifies the term is mapped to a tbx:Transaction (a subclass of prov:Activity). Provenance metadata is attached to this entity. The prov:Activity related to the responsible person or agent through prov:wasAssociatedWith; the relation to the responsible Agent is encoded via prov:wasGeneratedBy.

Converting terminological concepts and terms

Transforming terminological concepts

This is shown by the following RDF snippet, where the the subject field of the terminological concept is specified via the property subjectField:

Transforming TIGs

Our TBX example document has two language sets for English and German. In the lemon model, a lexicon is regarded as language-specific and as comprising lexical entries for a single language. Thus, in order to represent lexical entries in different languages, one lexicon per language needs to be created. In our example, as there are terms for English and German, two lexica need to be created. These lexica contain one lexical entry each, corresponding to the terms Zuständigkeit der Mitgliedstaaten and competence of the Member States. The English entry generated from lines 19--25 would look as follows:

Note that the entry specifies the reliability code (i.e. 3), the type of term (i.e. full form), the canonical form (i.e. competence of the member states), and the language (i.e. en). Each lexical entry is assumed to have a LexicalSense that represents the meaning of the entry. In this case the meaning is established by reference to the terminological concept :IATE_84.

We would generate a similar entry for German, which is identified as :Zust\%C3\%A4ndigkeit+\-der\-+Mitgliedstaaten-de and is an entry in the corresponding German lexicon. Note that both entries have a reference to :IATE_84 and are thus cross-lingual equivalents.

Transforming NTIGS

Components of a composite term are represented as constituents of the composite term. A component is linked to its corresponding lexical entry by way of the correspondsTo relation. In the example below, the lexical entry Zust%C3%A4ndigkeit+der+Mitgliedstaaten-de is linked to an object Zust%C3%A4ndigkeit+der+Mitgliedstaaten-de#ComponentList representing its decomposition via the property correspondsTo. This object Zust\%C3\%A4ndigkeit+der+Mitgliedstaaten-de#ComponentList is linked to its components via the property constituent. For each component, its part-of-speech and grammatical number (if applicable) is indicated. The decomposition of the German entry for Zuständigkeit der Mitgliedstaaten (lines 32-47 in the sample TBX document) is represented in RDF as indicated below:

Representing Transaction Information

Finally, we discuss how to represent provenance information, in particular provenance information as expressed via transaction elements in TBX. We rely on the PROV-O ontology for this, as this is the W3C recommended vocabulary to `represent and interchange provenance information generated in different systems and under different contexts´. Some provenance information is given on lines 48--53 of our TBX example document and from this we generate the following representation:

Indicate the number of terminological concepts in the resource

Indicates all the languages of terms included in the resource

Retrieve all the canonical forms for a given terminological concept in all languages, with indication of the language tag

Retrieve all terms by subject field